xref: /freebsd/sys/ufs/ufs/ufs_quota.c (revision e17f5b1d)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1990, 1993, 1995
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Robert Elz at The University of Melbourne.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)ufs_quota.c	8.5 (Berkeley) 5/20/95
35  */
36 
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39 
40 #include "opt_ffs.h"
41 
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/endian.h>
45 #include <sys/fcntl.h>
46 #include <sys/kernel.h>
47 #include <sys/lock.h>
48 #include <sys/malloc.h>
49 #include <sys/mount.h>
50 #include <sys/mutex.h>
51 #include <sys/namei.h>
52 #include <sys/priv.h>
53 #include <sys/proc.h>
54 #include <sys/socket.h>
55 #include <sys/stat.h>
56 #include <sys/sysctl.h>
57 #include <sys/vnode.h>
58 
59 #include <ufs/ufs/extattr.h>
60 #include <ufs/ufs/quota.h>
61 #include <ufs/ufs/inode.h>
62 #include <ufs/ufs/ufsmount.h>
63 #include <ufs/ufs/ufs_extern.h>
64 
65 CTASSERT(sizeof(struct dqblk64) == sizeof(struct dqhdr64));
66 
67 static int unprivileged_get_quota = 0;
68 SYSCTL_INT(_security_bsd, OID_AUTO, unprivileged_get_quota, CTLFLAG_RW,
69     &unprivileged_get_quota, 0,
70     "Unprivileged processes may retrieve quotas for other uids and gids");
71 
72 static MALLOC_DEFINE(M_DQUOT, "ufs_quota", "UFS quota entries");
73 
74 /*
75  * Quota name to error message mapping.
76  */
77 static char *quotatypes[] = INITQFNAMES;
78 
79 static int chkdqchg(struct inode *, ufs2_daddr_t, struct ucred *, int, int *);
80 static int chkiqchg(struct inode *, int, struct ucred *, int, int *);
81 static int dqopen(struct vnode *, struct ufsmount *, int);
82 static int dqget(struct vnode *,
83 	u_long, struct ufsmount *, int, struct dquot **);
84 static int dqsync(struct vnode *, struct dquot *);
85 static int dqflush(struct vnode *);
86 static int quotaoff1(struct thread *td, struct mount *mp, int type);
87 static int quotaoff_inchange(struct thread *td, struct mount *mp, int type);
88 
89 /* conversion functions - from_to() */
90 static void dqb32_dq(const struct dqblk32 *, struct dquot *);
91 static void dqb64_dq(const struct dqblk64 *, struct dquot *);
92 static void dq_dqb32(const struct dquot *, struct dqblk32 *);
93 static void dq_dqb64(const struct dquot *, struct dqblk64 *);
94 static void dqb32_dqb64(const struct dqblk32 *, struct dqblk64 *);
95 static void dqb64_dqb32(const struct dqblk64 *, struct dqblk32 *);
96 
97 #ifdef DIAGNOSTIC
98 static void dqref(struct dquot *);
99 static void chkdquot(struct inode *);
100 #endif
101 
102 /*
103  * Set up the quotas for an inode.
104  *
105  * This routine completely defines the semantics of quotas.
106  * If other criterion want to be used to establish quotas, the
107  * MAXQUOTAS value in quota.h should be increased, and the
108  * additional dquots set up here.
109  */
110 int
111 getinoquota(struct inode *ip)
112 {
113 	struct ufsmount *ump;
114 	struct vnode *vp;
115 	int error;
116 
117 	vp = ITOV(ip);
118 
119 	/*
120 	 * Disk quotas must be turned off for system files.  Currently
121 	 * snapshot and quota files.
122 	 */
123 	if ((vp->v_vflag & VV_SYSTEM) != 0)
124 		return (0);
125 	/*
126 	 * XXX: Turn off quotas for files with a negative UID or GID.
127 	 * This prevents the creation of 100GB+ quota files.
128 	 */
129 	if ((int)ip->i_uid < 0 || (int)ip->i_gid < 0)
130 		return (0);
131 	ump = VFSTOUFS(vp->v_mount);
132 	/*
133 	 * Set up the user quota based on file uid.
134 	 * EINVAL means that quotas are not enabled.
135 	 */
136 	if ((error =
137 		dqget(vp, ip->i_uid, ump, USRQUOTA, &ip->i_dquot[USRQUOTA])) &&
138 	    error != EINVAL)
139 		return (error);
140 	/*
141 	 * Set up the group quota based on file gid.
142 	 * EINVAL means that quotas are not enabled.
143 	 */
144 	if ((error =
145 		dqget(vp, ip->i_gid, ump, GRPQUOTA, &ip->i_dquot[GRPQUOTA])) &&
146 	    error != EINVAL)
147 		return (error);
148 	return (0);
149 }
150 
151 /*
152  * Update disk usage, and take corrective action.
153  */
154 int
155 chkdq(struct inode *ip, ufs2_daddr_t change, struct ucred *cred, int flags)
156 {
157 	struct dquot *dq;
158 	ufs2_daddr_t ncurblocks;
159 	struct vnode *vp = ITOV(ip);
160 	int i, error, warn, do_check;
161 
162 	MPASS(cred != NOCRED || (flags & FORCE) != 0);
163 	/*
164 	 * Disk quotas must be turned off for system files.  Currently
165 	 * snapshot and quota files.
166 	 */
167 	if ((vp->v_vflag & VV_SYSTEM) != 0)
168 		return (0);
169 	/*
170 	 * XXX: Turn off quotas for files with a negative UID or GID.
171 	 * This prevents the creation of 100GB+ quota files.
172 	 */
173 	if ((int)ip->i_uid < 0 || (int)ip->i_gid < 0)
174 		return (0);
175 #ifdef DIAGNOSTIC
176 	if ((flags & CHOWN) == 0)
177 		chkdquot(ip);
178 #endif
179 	if (change == 0)
180 		return (0);
181 	if (change < 0) {
182 		for (i = 0; i < MAXQUOTAS; i++) {
183 			if ((dq = ip->i_dquot[i]) == NODQUOT)
184 				continue;
185 			DQI_LOCK(dq);
186 			DQI_WAIT(dq, PINOD+1, "chkdq1");
187 			ncurblocks = dq->dq_curblocks + change;
188 			if (ncurblocks >= 0)
189 				dq->dq_curblocks = ncurblocks;
190 			else
191 				dq->dq_curblocks = 0;
192 			dq->dq_flags &= ~DQ_BLKS;
193 			dq->dq_flags |= DQ_MOD;
194 			DQI_UNLOCK(dq);
195 		}
196 		return (0);
197 	}
198 	if ((flags & FORCE) == 0 &&
199 	    priv_check_cred(cred, PRIV_VFS_EXCEEDQUOTA))
200 		do_check = 1;
201 	else
202 		do_check = 0;
203 	for (i = 0; i < MAXQUOTAS; i++) {
204 		if ((dq = ip->i_dquot[i]) == NODQUOT)
205 			continue;
206 		warn = 0;
207 		DQI_LOCK(dq);
208 		DQI_WAIT(dq, PINOD+1, "chkdq2");
209 		if (do_check) {
210 			error = chkdqchg(ip, change, cred, i, &warn);
211 			if (error) {
212 				/*
213 				 * Roll back user quota changes when
214 				 * group quota failed.
215 				 */
216 				while (i > 0) {
217 					--i;
218 					dq = ip->i_dquot[i];
219 					if (dq == NODQUOT)
220 						continue;
221 					DQI_LOCK(dq);
222 					DQI_WAIT(dq, PINOD+1, "chkdq3");
223 					ncurblocks = dq->dq_curblocks - change;
224 					if (ncurblocks >= 0)
225 						dq->dq_curblocks = ncurblocks;
226 					else
227 						dq->dq_curblocks = 0;
228 					dq->dq_flags &= ~DQ_BLKS;
229 					dq->dq_flags |= DQ_MOD;
230 					DQI_UNLOCK(dq);
231 				}
232 				return (error);
233 			}
234 		}
235 		/* Reset timer when crossing soft limit */
236 		if (dq->dq_curblocks + change >= dq->dq_bsoftlimit &&
237 		    dq->dq_curblocks < dq->dq_bsoftlimit)
238 			dq->dq_btime = time_second + ITOUMP(ip)->um_btime[i];
239 		dq->dq_curblocks += change;
240 		dq->dq_flags |= DQ_MOD;
241 		DQI_UNLOCK(dq);
242 		if (warn)
243 			uprintf("\n%s: warning, %s disk quota exceeded\n",
244 			    ITOVFS(ip)->mnt_stat.f_mntonname,
245 			    quotatypes[i]);
246 	}
247 	return (0);
248 }
249 
250 /*
251  * Check for a valid change to a users allocation.
252  * Issue an error message if appropriate.
253  */
254 static int
255 chkdqchg(struct inode *ip, ufs2_daddr_t change, struct ucred *cred,
256     int type, int *warn)
257 {
258 	struct dquot *dq = ip->i_dquot[type];
259 	ufs2_daddr_t ncurblocks = dq->dq_curblocks + change;
260 
261 	/*
262 	 * If user would exceed their hard limit, disallow space allocation.
263 	 */
264 	if (ncurblocks >= dq->dq_bhardlimit && dq->dq_bhardlimit) {
265 		if ((dq->dq_flags & DQ_BLKS) == 0 &&
266 		    ip->i_uid == cred->cr_uid) {
267 			dq->dq_flags |= DQ_BLKS;
268 			DQI_UNLOCK(dq);
269 			uprintf("\n%s: write failed, %s disk limit reached\n",
270 			    ITOVFS(ip)->mnt_stat.f_mntonname,
271 			    quotatypes[type]);
272 			return (EDQUOT);
273 		}
274 		DQI_UNLOCK(dq);
275 		return (EDQUOT);
276 	}
277 	/*
278 	 * If user is over their soft limit for too long, disallow space
279 	 * allocation. Reset time limit as they cross their soft limit.
280 	 */
281 	if (ncurblocks >= dq->dq_bsoftlimit && dq->dq_bsoftlimit) {
282 		if (dq->dq_curblocks < dq->dq_bsoftlimit) {
283 			dq->dq_btime = time_second + ITOUMP(ip)->um_btime[type];
284 			if (ip->i_uid == cred->cr_uid)
285 				*warn = 1;
286 			return (0);
287 		}
288 		if (time_second > dq->dq_btime) {
289 			if ((dq->dq_flags & DQ_BLKS) == 0 &&
290 			    ip->i_uid == cred->cr_uid) {
291 				dq->dq_flags |= DQ_BLKS;
292 				DQI_UNLOCK(dq);
293 				uprintf("\n%s: write failed, %s "
294 				    "disk quota exceeded for too long\n",
295 				    ITOVFS(ip)->mnt_stat.f_mntonname,
296 				    quotatypes[type]);
297 				return (EDQUOT);
298 			}
299 			DQI_UNLOCK(dq);
300 			return (EDQUOT);
301 		}
302 	}
303 	return (0);
304 }
305 
306 /*
307  * Check the inode limit, applying corrective action.
308  */
309 int
310 chkiq(struct inode *ip, int change, struct ucred *cred, int flags)
311 {
312 	struct dquot *dq;
313 	int i, error, warn, do_check;
314 
315 	MPASS(cred != NOCRED || (flags & FORCE) != 0);
316 #ifdef DIAGNOSTIC
317 	if ((flags & CHOWN) == 0)
318 		chkdquot(ip);
319 #endif
320 	if (change == 0)
321 		return (0);
322 	if (change < 0) {
323 		for (i = 0; i < MAXQUOTAS; i++) {
324 			if ((dq = ip->i_dquot[i]) == NODQUOT)
325 				continue;
326 			DQI_LOCK(dq);
327 			DQI_WAIT(dq, PINOD+1, "chkiq1");
328 			if (dq->dq_curinodes >= -change)
329 				dq->dq_curinodes += change;
330 			else
331 				dq->dq_curinodes = 0;
332 			dq->dq_flags &= ~DQ_INODS;
333 			dq->dq_flags |= DQ_MOD;
334 			DQI_UNLOCK(dq);
335 		}
336 		return (0);
337 	}
338 	if ((flags & FORCE) == 0 &&
339 	    priv_check_cred(cred, PRIV_VFS_EXCEEDQUOTA))
340 		do_check = 1;
341 	else
342 		do_check = 0;
343 	for (i = 0; i < MAXQUOTAS; i++) {
344 		if ((dq = ip->i_dquot[i]) == NODQUOT)
345 			continue;
346 		warn = 0;
347 		DQI_LOCK(dq);
348 		DQI_WAIT(dq, PINOD+1, "chkiq2");
349 		if (do_check) {
350 			error = chkiqchg(ip, change, cred, i, &warn);
351 			if (error) {
352 				/*
353 				 * Roll back user quota changes when
354 				 * group quota failed.
355 				 */
356 				while (i > 0) {
357 					--i;
358 					dq = ip->i_dquot[i];
359 					if (dq == NODQUOT)
360 						continue;
361 					DQI_LOCK(dq);
362 					DQI_WAIT(dq, PINOD+1, "chkiq3");
363 					if (dq->dq_curinodes >= change)
364 						dq->dq_curinodes -= change;
365 					else
366 						dq->dq_curinodes = 0;
367 					dq->dq_flags &= ~DQ_INODS;
368 					dq->dq_flags |= DQ_MOD;
369 					DQI_UNLOCK(dq);
370 				}
371 				return (error);
372 			}
373 		}
374 		/* Reset timer when crossing soft limit */
375 		if (dq->dq_curinodes + change >= dq->dq_isoftlimit &&
376 		    dq->dq_curinodes < dq->dq_isoftlimit)
377 			dq->dq_itime = time_second + ITOUMP(ip)->um_itime[i];
378 		dq->dq_curinodes += change;
379 		dq->dq_flags |= DQ_MOD;
380 		DQI_UNLOCK(dq);
381 		if (warn)
382 			uprintf("\n%s: warning, %s inode quota exceeded\n",
383 			    ITOVFS(ip)->mnt_stat.f_mntonname,
384 			    quotatypes[i]);
385 	}
386 	return (0);
387 }
388 
389 /*
390  * Check for a valid change to a users allocation.
391  * Issue an error message if appropriate.
392  */
393 static int
394 chkiqchg(struct inode *ip, int change, struct ucred *cred, int type, int *warn)
395 {
396 	struct dquot *dq = ip->i_dquot[type];
397 	ino_t ncurinodes = dq->dq_curinodes + change;
398 
399 	/*
400 	 * If user would exceed their hard limit, disallow inode allocation.
401 	 */
402 	if (ncurinodes >= dq->dq_ihardlimit && dq->dq_ihardlimit) {
403 		if ((dq->dq_flags & DQ_INODS) == 0 &&
404 		    ip->i_uid == cred->cr_uid) {
405 			dq->dq_flags |= DQ_INODS;
406 			DQI_UNLOCK(dq);
407 			uprintf("\n%s: write failed, %s inode limit reached\n",
408 			    ITOVFS(ip)->mnt_stat.f_mntonname,
409 			    quotatypes[type]);
410 			return (EDQUOT);
411 		}
412 		DQI_UNLOCK(dq);
413 		return (EDQUOT);
414 	}
415 	/*
416 	 * If user is over their soft limit for too long, disallow inode
417 	 * allocation. Reset time limit as they cross their soft limit.
418 	 */
419 	if (ncurinodes >= dq->dq_isoftlimit && dq->dq_isoftlimit) {
420 		if (dq->dq_curinodes < dq->dq_isoftlimit) {
421 			dq->dq_itime = time_second + ITOUMP(ip)->um_itime[type];
422 			if (ip->i_uid == cred->cr_uid)
423 				*warn = 1;
424 			return (0);
425 		}
426 		if (time_second > dq->dq_itime) {
427 			if ((dq->dq_flags & DQ_INODS) == 0 &&
428 			    ip->i_uid == cred->cr_uid) {
429 				dq->dq_flags |= DQ_INODS;
430 				DQI_UNLOCK(dq);
431 				uprintf("\n%s: write failed, %s "
432 				    "inode quota exceeded for too long\n",
433 				    ITOVFS(ip)->mnt_stat.f_mntonname,
434 				    quotatypes[type]);
435 				return (EDQUOT);
436 			}
437 			DQI_UNLOCK(dq);
438 			return (EDQUOT);
439 		}
440 	}
441 	return (0);
442 }
443 
444 #ifdef DIAGNOSTIC
445 /*
446  * On filesystems with quotas enabled, it is an error for a file to change
447  * size and not to have a dquot structure associated with it.
448  */
449 static void
450 chkdquot(struct inode *ip)
451 {
452 	struct ufsmount *ump;
453 	struct vnode *vp;
454 	int i;
455 
456 	ump = ITOUMP(ip);
457 	vp = ITOV(ip);
458 
459 	/*
460 	 * Disk quotas must be turned off for system files.  Currently
461 	 * these are snapshots and quota files.
462 	 */
463 	if ((vp->v_vflag & VV_SYSTEM) != 0)
464 		return;
465 	/*
466 	 * XXX: Turn off quotas for files with a negative UID or GID.
467 	 * This prevents the creation of 100GB+ quota files.
468 	 */
469 	if ((int)ip->i_uid < 0 || (int)ip->i_gid < 0)
470 		return;
471 
472 	UFS_LOCK(ump);
473 	for (i = 0; i < MAXQUOTAS; i++) {
474 		if (ump->um_quotas[i] == NULLVP ||
475 		    (ump->um_qflags[i] & (QTF_OPENING|QTF_CLOSING)))
476 			continue;
477 		if (ip->i_dquot[i] == NODQUOT) {
478 			UFS_UNLOCK(ump);
479 			vn_printf(ITOV(ip), "chkdquot: missing dquot ");
480 			panic("chkdquot: missing dquot");
481 		}
482 	}
483 	UFS_UNLOCK(ump);
484 }
485 #endif
486 
487 /*
488  * Code to process quotactl commands.
489  */
490 
491 /*
492  * Q_QUOTAON - set up a quota file for a particular filesystem.
493  */
494 int
495 quotaon(struct thread *td, struct mount *mp, int type, void *fname)
496 {
497 	struct ufsmount *ump;
498 	struct vnode *vp, **vpp;
499 	struct vnode *mvp;
500 	struct dquot *dq;
501 	int error, flags;
502 	struct nameidata nd;
503 
504 	error = priv_check(td, PRIV_UFS_QUOTAON);
505 	if (error != 0) {
506 		vfs_unbusy(mp);
507 		return (error);
508 	}
509 
510 	if ((mp->mnt_flag & MNT_RDONLY) != 0) {
511 		vfs_unbusy(mp);
512 		return (EROFS);
513 	}
514 
515 	ump = VFSTOUFS(mp);
516 	dq = NODQUOT;
517 
518 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, fname, td);
519 	flags = FREAD | FWRITE;
520 	vfs_ref(mp);
521 	vfs_unbusy(mp);
522 	error = vn_open(&nd, &flags, 0, NULL);
523 	if (error != 0) {
524 		vfs_rel(mp);
525 		return (error);
526 	}
527 	NDFREE(&nd, NDF_ONLY_PNBUF);
528 	vp = nd.ni_vp;
529 	error = vfs_busy(mp, MBF_NOWAIT);
530 	vfs_rel(mp);
531 	if (error == 0) {
532 		if (vp->v_type != VREG) {
533 			error = EACCES;
534 			vfs_unbusy(mp);
535 		}
536 	}
537 	if (error != 0) {
538 		VOP_UNLOCK(vp);
539 		(void) vn_close(vp, FREAD|FWRITE, td->td_ucred, td);
540 		return (error);
541 	}
542 
543 	UFS_LOCK(ump);
544 	if ((ump->um_qflags[type] & (QTF_OPENING|QTF_CLOSING)) != 0) {
545 		UFS_UNLOCK(ump);
546 		VOP_UNLOCK(vp);
547 		(void) vn_close(vp, FREAD|FWRITE, td->td_ucred, td);
548 		vfs_unbusy(mp);
549 		return (EALREADY);
550 	}
551 	ump->um_qflags[type] |= QTF_OPENING|QTF_CLOSING;
552 	UFS_UNLOCK(ump);
553 	if ((error = dqopen(vp, ump, type)) != 0) {
554 		VOP_UNLOCK(vp);
555 		UFS_LOCK(ump);
556 		ump->um_qflags[type] &= ~(QTF_OPENING|QTF_CLOSING);
557 		UFS_UNLOCK(ump);
558 		(void) vn_close(vp, FREAD|FWRITE, td->td_ucred, td);
559 		vfs_unbusy(mp);
560 		return (error);
561 	}
562 	VOP_UNLOCK(vp);
563 	MNT_ILOCK(mp);
564 	mp->mnt_flag |= MNT_QUOTA;
565 	MNT_IUNLOCK(mp);
566 
567 	vpp = &ump->um_quotas[type];
568 	if (*vpp != vp)
569 		quotaoff1(td, mp, type);
570 
571 	/*
572 	 * When the directory vnode containing the quota file is
573 	 * inactivated, due to the shared lookup of the quota file
574 	 * vput()ing the dvp, the qsyncvp() call for the containing
575 	 * directory would try to acquire the quota lock exclusive.
576 	 * At the same time, lookup already locked the quota vnode
577 	 * shared.  Mark the quota vnode lock as allowing recursion
578 	 * and automatically converting shared locks to exclusive.
579 	 *
580 	 * Also mark quota vnode as system.
581 	 */
582 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
583 	vp->v_vflag |= VV_SYSTEM;
584 	VN_LOCK_AREC(vp);
585 	VN_LOCK_DSHARE(vp);
586 	VOP_UNLOCK(vp);
587 	*vpp = vp;
588 	/*
589 	 * Save the credential of the process that turned on quotas.
590 	 * Set up the time limits for this quota.
591 	 */
592 	ump->um_cred[type] = crhold(td->td_ucred);
593 	ump->um_btime[type] = MAX_DQ_TIME;
594 	ump->um_itime[type] = MAX_IQ_TIME;
595 	if (dqget(NULLVP, 0, ump, type, &dq) == 0) {
596 		if (dq->dq_btime > 0)
597 			ump->um_btime[type] = dq->dq_btime;
598 		if (dq->dq_itime > 0)
599 			ump->um_itime[type] = dq->dq_itime;
600 		dqrele(NULLVP, dq);
601 	}
602 	/*
603 	 * Allow the getdq from getinoquota below to read the quota
604 	 * from file.
605 	 */
606 	UFS_LOCK(ump);
607 	ump->um_qflags[type] &= ~QTF_CLOSING;
608 	UFS_UNLOCK(ump);
609 	/*
610 	 * Search vnodes associated with this mount point,
611 	 * adding references to quota file being opened.
612 	 * NB: only need to add dquot's for inodes being modified.
613 	 */
614 again:
615 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
616 		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) {
617 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
618 			goto again;
619 		}
620 		if (vp->v_type == VNON || vp->v_writecount <= 0) {
621 			VOP_UNLOCK(vp);
622 			vrele(vp);
623 			continue;
624 		}
625 		error = getinoquota(VTOI(vp));
626 		VOP_UNLOCK(vp);
627 		vrele(vp);
628 		if (error) {
629 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
630 			break;
631 		}
632 	}
633 
634         if (error)
635 		quotaoff_inchange(td, mp, type);
636 	UFS_LOCK(ump);
637 	ump->um_qflags[type] &= ~QTF_OPENING;
638 	KASSERT((ump->um_qflags[type] & QTF_CLOSING) == 0,
639 		("quotaon: leaking flags"));
640 	UFS_UNLOCK(ump);
641 
642 	vfs_unbusy(mp);
643 	return (error);
644 }
645 
646 /*
647  * Main code to turn off disk quotas for a filesystem. Does not change
648  * flags.
649  */
650 static int
651 quotaoff1(struct thread *td, struct mount *mp, int type)
652 {
653 	struct vnode *vp;
654 	struct vnode *qvp, *mvp;
655 	struct ufsmount *ump;
656 	struct dquot *dq;
657 	struct inode *ip;
658 	struct ucred *cr;
659 	int error;
660 
661 	ump = VFSTOUFS(mp);
662 
663 	UFS_LOCK(ump);
664 	KASSERT((ump->um_qflags[type] & QTF_CLOSING) != 0,
665 		("quotaoff1: flags are invalid"));
666 	if ((qvp = ump->um_quotas[type]) == NULLVP) {
667 		UFS_UNLOCK(ump);
668 		return (0);
669 	}
670 	cr = ump->um_cred[type];
671 	UFS_UNLOCK(ump);
672 
673 	/*
674 	 * Search vnodes associated with this mount point,
675 	 * deleting any references to quota file being closed.
676 	 */
677 again:
678 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
679 		if (vp->v_type == VNON) {
680 			VI_UNLOCK(vp);
681 			continue;
682 		}
683 		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) {
684 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
685 			goto again;
686 		}
687 		ip = VTOI(vp);
688 		dq = ip->i_dquot[type];
689 		ip->i_dquot[type] = NODQUOT;
690 		dqrele(vp, dq);
691 		VOP_UNLOCK(vp);
692 		vrele(vp);
693 	}
694 
695 	error = dqflush(qvp);
696 	if (error != 0)
697 		return (error);
698 
699 	/*
700 	 * Clear um_quotas before closing the quota vnode to prevent
701 	 * access to the closed vnode from dqget/dqsync
702 	 */
703 	UFS_LOCK(ump);
704 	ump->um_quotas[type] = NULLVP;
705 	ump->um_cred[type] = NOCRED;
706 	UFS_UNLOCK(ump);
707 
708 	vn_lock(qvp, LK_EXCLUSIVE | LK_RETRY);
709 	qvp->v_vflag &= ~VV_SYSTEM;
710 	VOP_UNLOCK(qvp);
711 	error = vn_close(qvp, FREAD|FWRITE, td->td_ucred, td);
712 	crfree(cr);
713 
714 	return (error);
715 }
716 
717 static int
718 quotaoff_inchange1(struct thread *td, struct mount *mp, int type)
719 {
720 	int error;
721 	bool need_resume;
722 
723 	/*
724 	 * mp is already suspended on unmount.  If not, suspend it, to
725 	 * avoid the situation where quotaoff operation eventually
726 	 * failing due to SU structures still keeping references on
727 	 * dquots, but vnode's references are already clean.  This
728 	 * would cause quota accounting leak and asserts otherwise.
729 	 * Note that the thread has already called vn_start_write().
730 	 */
731 	if (mp->mnt_susp_owner == td) {
732 		need_resume = false;
733 	} else {
734 		error = vfs_write_suspend_umnt(mp);
735 		if (error != 0)
736 			return (error);
737 		need_resume = true;
738 	}
739 	error = quotaoff1(td, mp, type);
740 	if (need_resume)
741 		vfs_write_resume(mp, VR_START_WRITE);
742 	return (error);
743 }
744 
745 /*
746  * Turns off quotas, assumes that ump->um_qflags are already checked
747  * and QTF_CLOSING is set to indicate operation in progress. Fixes
748  * ump->um_qflags and mp->mnt_flag after.
749  */
750 int
751 quotaoff_inchange(struct thread *td, struct mount *mp, int type)
752 {
753 	struct ufsmount *ump;
754 	int error, i;
755 
756 	error = quotaoff_inchange1(td, mp, type);
757 
758 	ump = VFSTOUFS(mp);
759 	UFS_LOCK(ump);
760 	ump->um_qflags[type] &= ~QTF_CLOSING;
761 	for (i = 0; i < MAXQUOTAS; i++)
762 		if (ump->um_quotas[i] != NULLVP)
763 			break;
764 	if (i == MAXQUOTAS) {
765 		MNT_ILOCK(mp);
766 		mp->mnt_flag &= ~MNT_QUOTA;
767 		MNT_IUNLOCK(mp);
768 	}
769 	UFS_UNLOCK(ump);
770 	return (error);
771 }
772 
773 /*
774  * Q_QUOTAOFF - turn off disk quotas for a filesystem.
775  */
776 int
777 quotaoff(struct thread *td, struct mount *mp, int type)
778 {
779 	struct ufsmount *ump;
780 	int error;
781 
782 	error = priv_check(td, PRIV_UFS_QUOTAOFF);
783 	if (error)
784 		return (error);
785 
786 	ump = VFSTOUFS(mp);
787 	UFS_LOCK(ump);
788 	if ((ump->um_qflags[type] & (QTF_OPENING|QTF_CLOSING)) != 0) {
789 		UFS_UNLOCK(ump);
790 		return (EALREADY);
791 	}
792 	ump->um_qflags[type] |= QTF_CLOSING;
793 	UFS_UNLOCK(ump);
794 
795 	return (quotaoff_inchange(td, mp, type));
796 }
797 
798 /*
799  * Q_GETQUOTA - return current values in a dqblk structure.
800  */
801 static int
802 _getquota(struct thread *td, struct mount *mp, u_long id, int type,
803     struct dqblk64 *dqb)
804 {
805 	struct dquot *dq;
806 	int error;
807 
808 	switch (type) {
809 	case USRQUOTA:
810 		if ((td->td_ucred->cr_uid != id) && !unprivileged_get_quota) {
811 			error = priv_check(td, PRIV_VFS_GETQUOTA);
812 			if (error)
813 				return (error);
814 		}
815 		break;
816 
817 	case GRPQUOTA:
818 		if (!groupmember(id, td->td_ucred) &&
819 		    !unprivileged_get_quota) {
820 			error = priv_check(td, PRIV_VFS_GETQUOTA);
821 			if (error)
822 				return (error);
823 		}
824 		break;
825 
826 	default:
827 		return (EINVAL);
828 	}
829 
830 	dq = NODQUOT;
831 	error = dqget(NULLVP, id, VFSTOUFS(mp), type, &dq);
832 	if (error)
833 		return (error);
834 	*dqb = dq->dq_dqb;
835 	dqrele(NULLVP, dq);
836 	return (error);
837 }
838 
839 /*
840  * Q_SETQUOTA - assign an entire dqblk structure.
841  */
842 static int
843 _setquota(struct thread *td, struct mount *mp, u_long id, int type,
844     struct dqblk64 *dqb)
845 {
846 	struct dquot *dq;
847 	struct dquot *ndq;
848 	struct ufsmount *ump;
849 	struct dqblk64 newlim;
850 	int error;
851 
852 	error = priv_check(td, PRIV_VFS_SETQUOTA);
853 	if (error)
854 		return (error);
855 
856 	newlim = *dqb;
857 
858 	ndq = NODQUOT;
859 	ump = VFSTOUFS(mp);
860 
861 	error = dqget(NULLVP, id, ump, type, &ndq);
862 	if (error)
863 		return (error);
864 	dq = ndq;
865 	DQI_LOCK(dq);
866 	DQI_WAIT(dq, PINOD+1, "setqta");
867 	/*
868 	 * Copy all but the current values.
869 	 * Reset time limit if previously had no soft limit or were
870 	 * under it, but now have a soft limit and are over it.
871 	 */
872 	newlim.dqb_curblocks = dq->dq_curblocks;
873 	newlim.dqb_curinodes = dq->dq_curinodes;
874 	if (dq->dq_id != 0) {
875 		newlim.dqb_btime = dq->dq_btime;
876 		newlim.dqb_itime = dq->dq_itime;
877 	}
878 	if (newlim.dqb_bsoftlimit &&
879 	    dq->dq_curblocks >= newlim.dqb_bsoftlimit &&
880 	    (dq->dq_bsoftlimit == 0 || dq->dq_curblocks < dq->dq_bsoftlimit))
881 		newlim.dqb_btime = time_second + ump->um_btime[type];
882 	if (newlim.dqb_isoftlimit &&
883 	    dq->dq_curinodes >= newlim.dqb_isoftlimit &&
884 	    (dq->dq_isoftlimit == 0 || dq->dq_curinodes < dq->dq_isoftlimit))
885 		newlim.dqb_itime = time_second + ump->um_itime[type];
886 	dq->dq_dqb = newlim;
887 	if (dq->dq_curblocks < dq->dq_bsoftlimit)
888 		dq->dq_flags &= ~DQ_BLKS;
889 	if (dq->dq_curinodes < dq->dq_isoftlimit)
890 		dq->dq_flags &= ~DQ_INODS;
891 	if (dq->dq_isoftlimit == 0 && dq->dq_bsoftlimit == 0 &&
892 	    dq->dq_ihardlimit == 0 && dq->dq_bhardlimit == 0)
893 		dq->dq_flags |= DQ_FAKE;
894 	else
895 		dq->dq_flags &= ~DQ_FAKE;
896 	dq->dq_flags |= DQ_MOD;
897 	DQI_UNLOCK(dq);
898 	dqrele(NULLVP, dq);
899 	return (0);
900 }
901 
902 /*
903  * Q_SETUSE - set current inode and block usage.
904  */
905 static int
906 _setuse(struct thread *td, struct mount *mp, u_long id, int type,
907     struct dqblk64 *dqb)
908 {
909 	struct dquot *dq;
910 	struct ufsmount *ump;
911 	struct dquot *ndq;
912 	struct dqblk64 usage;
913 	int error;
914 
915 	error = priv_check(td, PRIV_UFS_SETUSE);
916 	if (error)
917 		return (error);
918 
919 	usage = *dqb;
920 
921 	ump = VFSTOUFS(mp);
922 	ndq = NODQUOT;
923 
924 	error = dqget(NULLVP, id, ump, type, &ndq);
925 	if (error)
926 		return (error);
927 	dq = ndq;
928 	DQI_LOCK(dq);
929 	DQI_WAIT(dq, PINOD+1, "setuse");
930 	/*
931 	 * Reset time limit if have a soft limit and were
932 	 * previously under it, but are now over it.
933 	 */
934 	if (dq->dq_bsoftlimit && dq->dq_curblocks < dq->dq_bsoftlimit &&
935 	    usage.dqb_curblocks >= dq->dq_bsoftlimit)
936 		dq->dq_btime = time_second + ump->um_btime[type];
937 	if (dq->dq_isoftlimit && dq->dq_curinodes < dq->dq_isoftlimit &&
938 	    usage.dqb_curinodes >= dq->dq_isoftlimit)
939 		dq->dq_itime = time_second + ump->um_itime[type];
940 	dq->dq_curblocks = usage.dqb_curblocks;
941 	dq->dq_curinodes = usage.dqb_curinodes;
942 	if (dq->dq_curblocks < dq->dq_bsoftlimit)
943 		dq->dq_flags &= ~DQ_BLKS;
944 	if (dq->dq_curinodes < dq->dq_isoftlimit)
945 		dq->dq_flags &= ~DQ_INODS;
946 	dq->dq_flags |= DQ_MOD;
947 	DQI_UNLOCK(dq);
948 	dqrele(NULLVP, dq);
949 	return (0);
950 }
951 
952 int
953 getquota32(struct thread *td, struct mount *mp, u_long id, int type, void *addr)
954 {
955 	struct dqblk32 dqb32;
956 	struct dqblk64 dqb64;
957 	int error;
958 
959 	error = _getquota(td, mp, id, type, &dqb64);
960 	if (error)
961 		return (error);
962 	dqb64_dqb32(&dqb64, &dqb32);
963 	error = copyout(&dqb32, addr, sizeof(dqb32));
964 	return (error);
965 }
966 
967 int
968 setquota32(struct thread *td, struct mount *mp, u_long id, int type, void *addr)
969 {
970 	struct dqblk32 dqb32;
971 	struct dqblk64 dqb64;
972 	int error;
973 
974 	error = copyin(addr, &dqb32, sizeof(dqb32));
975 	if (error)
976 		return (error);
977 	dqb32_dqb64(&dqb32, &dqb64);
978 	error = _setquota(td, mp, id, type, &dqb64);
979 	return (error);
980 }
981 
982 int
983 setuse32(struct thread *td, struct mount *mp, u_long id, int type, void *addr)
984 {
985 	struct dqblk32 dqb32;
986 	struct dqblk64 dqb64;
987 	int error;
988 
989 	error = copyin(addr, &dqb32, sizeof(dqb32));
990 	if (error)
991 		return (error);
992 	dqb32_dqb64(&dqb32, &dqb64);
993 	error = _setuse(td, mp, id, type, &dqb64);
994 	return (error);
995 }
996 
997 int
998 getquota(struct thread *td, struct mount *mp, u_long id, int type, void *addr)
999 {
1000 	struct dqblk64 dqb64;
1001 	int error;
1002 
1003 	error = _getquota(td, mp, id, type, &dqb64);
1004 	if (error)
1005 		return (error);
1006 	error = copyout(&dqb64, addr, sizeof(dqb64));
1007 	return (error);
1008 }
1009 
1010 int
1011 setquota(struct thread *td, struct mount *mp, u_long id, int type, void *addr)
1012 {
1013 	struct dqblk64 dqb64;
1014 	int error;
1015 
1016 	error = copyin(addr, &dqb64, sizeof(dqb64));
1017 	if (error)
1018 		return (error);
1019 	error = _setquota(td, mp, id, type, &dqb64);
1020 	return (error);
1021 }
1022 
1023 int
1024 setuse(struct thread *td, struct mount *mp, u_long id, int type, void *addr)
1025 {
1026 	struct dqblk64 dqb64;
1027 	int error;
1028 
1029 	error = copyin(addr, &dqb64, sizeof(dqb64));
1030 	if (error)
1031 		return (error);
1032 	error = _setuse(td, mp, id, type, &dqb64);
1033 	return (error);
1034 }
1035 
1036 /*
1037  * Q_GETQUOTASIZE - get bit-size of quota file fields
1038  */
1039 int
1040 getquotasize(struct thread *td, struct mount *mp, u_long id, int type,
1041     void *sizep)
1042 {
1043 	struct ufsmount *ump = VFSTOUFS(mp);
1044 	int bitsize;
1045 
1046 	UFS_LOCK(ump);
1047 	if (ump->um_quotas[type] == NULLVP ||
1048 	    (ump->um_qflags[type] & QTF_CLOSING)) {
1049 		UFS_UNLOCK(ump);
1050 		return (EINVAL);
1051 	}
1052 	if ((ump->um_qflags[type] & QTF_64BIT) != 0)
1053 		bitsize = 64;
1054 	else
1055 		bitsize = 32;
1056 	UFS_UNLOCK(ump);
1057 	return (copyout(&bitsize, sizep, sizeof(int)));
1058 }
1059 
1060 /*
1061  * Q_SYNC - sync quota files to disk.
1062  */
1063 int
1064 qsync(struct mount *mp)
1065 {
1066 	struct ufsmount *ump = VFSTOUFS(mp);
1067 	struct thread *td = curthread;		/* XXX */
1068 	struct vnode *vp, *mvp;
1069 	struct dquot *dq;
1070 	int i, error;
1071 
1072 	/*
1073 	 * Check if the mount point has any quotas.
1074 	 * If not, simply return.
1075 	 */
1076 	for (i = 0; i < MAXQUOTAS; i++)
1077 		if (ump->um_quotas[i] != NULLVP)
1078 			break;
1079 	if (i == MAXQUOTAS)
1080 		return (0);
1081 	/*
1082 	 * Search vnodes associated with this mount point,
1083 	 * synchronizing any modified dquot structures.
1084 	 */
1085 again:
1086 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1087 		if (vp->v_type == VNON) {
1088 			VI_UNLOCK(vp);
1089 			continue;
1090 		}
1091 		error = vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td);
1092 		if (error) {
1093 			if (error == ENOENT) {
1094 				MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1095 				goto again;
1096 			}
1097 			continue;
1098 		}
1099 		for (i = 0; i < MAXQUOTAS; i++) {
1100 			dq = VTOI(vp)->i_dquot[i];
1101 			if (dq != NODQUOT)
1102 				dqsync(vp, dq);
1103 		}
1104 		vput(vp);
1105 	}
1106 	return (0);
1107 }
1108 
1109 /*
1110  * Sync quota file for given vnode to disk.
1111  */
1112 int
1113 qsyncvp(struct vnode *vp)
1114 {
1115 	struct ufsmount *ump = VFSTOUFS(vp->v_mount);
1116 	struct dquot *dq;
1117 	int i;
1118 
1119 	/*
1120 	 * Check if the mount point has any quotas.
1121 	 * If not, simply return.
1122 	 */
1123 	for (i = 0; i < MAXQUOTAS; i++)
1124 		if (ump->um_quotas[i] != NULLVP)
1125 			break;
1126 	if (i == MAXQUOTAS)
1127 		return (0);
1128 	/*
1129 	 * Search quotas associated with this vnode
1130 	 * synchronizing any modified dquot structures.
1131 	 */
1132 	for (i = 0; i < MAXQUOTAS; i++) {
1133 		dq = VTOI(vp)->i_dquot[i];
1134 		if (dq != NODQUOT)
1135 			dqsync(vp, dq);
1136 	}
1137 	return (0);
1138 }
1139 
1140 /*
1141  * Code pertaining to management of the in-core dquot data structures.
1142  */
1143 #define DQHASH(dqvp, id) \
1144 	(&dqhashtbl[((((intptr_t)(dqvp)) >> 8) + id) & dqhash])
1145 static LIST_HEAD(dqhash, dquot) *dqhashtbl;
1146 static u_long dqhash;
1147 
1148 /*
1149  * Dquot free list.
1150  */
1151 #define	DQUOTINC	5	/* minimum free dquots desired */
1152 static TAILQ_HEAD(dqfreelist, dquot) dqfreelist;
1153 static long numdquot, desireddquot = DQUOTINC;
1154 
1155 /*
1156  * Lock to protect quota hash, dq free list and dq_cnt ref counters of
1157  * _all_ dqs.
1158  */
1159 struct mtx dqhlock;
1160 
1161 #define	DQH_LOCK()	mtx_lock(&dqhlock)
1162 #define	DQH_UNLOCK()	mtx_unlock(&dqhlock)
1163 
1164 static struct dquot *dqhashfind(struct dqhash *dqh, u_long id,
1165 	struct vnode *dqvp);
1166 
1167 /*
1168  * Initialize the quota system.
1169  */
1170 void
1171 dqinit(void)
1172 {
1173 
1174 	mtx_init(&dqhlock, "dqhlock", NULL, MTX_DEF);
1175 	dqhashtbl = hashinit(desiredvnodes, M_DQUOT, &dqhash);
1176 	TAILQ_INIT(&dqfreelist);
1177 }
1178 
1179 /*
1180  * Shut down the quota system.
1181  */
1182 void
1183 dquninit(void)
1184 {
1185 	struct dquot *dq;
1186 
1187 	hashdestroy(dqhashtbl, M_DQUOT, dqhash);
1188 	while ((dq = TAILQ_FIRST(&dqfreelist)) != NULL) {
1189 		TAILQ_REMOVE(&dqfreelist, dq, dq_freelist);
1190 		mtx_destroy(&dq->dq_lock);
1191 		free(dq, M_DQUOT);
1192 	}
1193 	mtx_destroy(&dqhlock);
1194 }
1195 
1196 static struct dquot *
1197 dqhashfind(struct dqhash *dqh, u_long id, struct vnode *dqvp)
1198 {
1199 	struct dquot *dq;
1200 
1201 	mtx_assert(&dqhlock, MA_OWNED);
1202 	LIST_FOREACH(dq, dqh, dq_hash) {
1203 		if (dq->dq_id != id ||
1204 		    dq->dq_ump->um_quotas[dq->dq_type] != dqvp)
1205 			continue;
1206 		/*
1207 		 * Cache hit with no references.  Take
1208 		 * the structure off the free list.
1209 		 */
1210 		if (dq->dq_cnt == 0)
1211 			TAILQ_REMOVE(&dqfreelist, dq, dq_freelist);
1212 		DQREF(dq);
1213 		return (dq);
1214 	}
1215 	return (NODQUOT);
1216 }
1217 
1218 /*
1219  * Determine the quota file type.
1220  *
1221  * A 32-bit quota file is simply an array of struct dqblk32.
1222  *
1223  * A 64-bit quota file is a struct dqhdr64 followed by an array of struct
1224  * dqblk64.  The header contains various magic bits which allow us to be
1225  * reasonably confident that it is indeeda 64-bit quota file and not just
1226  * a 32-bit quota file that just happens to "look right".
1227  *
1228  */
1229 static int
1230 dqopen(struct vnode *vp, struct ufsmount *ump, int type)
1231 {
1232 	struct dqhdr64 dqh;
1233 	struct iovec aiov;
1234 	struct uio auio;
1235 	int error;
1236 
1237 	ASSERT_VOP_LOCKED(vp, "dqopen");
1238 	auio.uio_iov = &aiov;
1239 	auio.uio_iovcnt = 1;
1240 	aiov.iov_base = &dqh;
1241 	aiov.iov_len = sizeof(dqh);
1242 	auio.uio_resid = sizeof(dqh);
1243 	auio.uio_offset = 0;
1244 	auio.uio_segflg = UIO_SYSSPACE;
1245 	auio.uio_rw = UIO_READ;
1246 	auio.uio_td = (struct thread *)0;
1247 	error = VOP_READ(vp, &auio, 0, ump->um_cred[type]);
1248 
1249 	if (error != 0)
1250 		return (error);
1251 	if (auio.uio_resid > 0) {
1252 		/* assume 32 bits */
1253 		return (0);
1254 	}
1255 
1256 	UFS_LOCK(ump);
1257 	if (strcmp(dqh.dqh_magic, Q_DQHDR64_MAGIC) == 0 &&
1258 	    be32toh(dqh.dqh_version) == Q_DQHDR64_VERSION &&
1259 	    be32toh(dqh.dqh_hdrlen) == (uint32_t)sizeof(struct dqhdr64) &&
1260 	    be32toh(dqh.dqh_reclen) == (uint32_t)sizeof(struct dqblk64)) {
1261 		/* XXX: what if the magic matches, but the sizes are wrong? */
1262 		ump->um_qflags[type] |= QTF_64BIT;
1263 	} else {
1264 		ump->um_qflags[type] &= ~QTF_64BIT;
1265 	}
1266 	UFS_UNLOCK(ump);
1267 
1268 	return (0);
1269 }
1270 
1271 /*
1272  * Obtain a dquot structure for the specified identifier and quota file
1273  * reading the information from the file if necessary.
1274  */
1275 static int
1276 dqget(struct vnode *vp, u_long id, struct ufsmount *ump, int type,
1277     struct dquot **dqp)
1278 {
1279 	uint8_t buf[sizeof(struct dqblk64)];
1280 	off_t base, recsize;
1281 	struct dquot *dq, *dq1;
1282 	struct dqhash *dqh;
1283 	struct vnode *dqvp;
1284 	struct iovec aiov;
1285 	struct uio auio;
1286 	int dqvplocked, error;
1287 
1288 #ifdef DEBUG_VFS_LOCKS
1289 	if (vp != NULLVP)
1290 		ASSERT_VOP_ELOCKED(vp, "dqget");
1291 #endif
1292 
1293 	if (vp != NULLVP && *dqp != NODQUOT) {
1294 		return (0);
1295 	}
1296 
1297 	/* XXX: Disallow negative id values to prevent the
1298 	* creation of 100GB+ quota data files.
1299 	*/
1300 	if ((int)id < 0)
1301 		return (EINVAL);
1302 
1303 	UFS_LOCK(ump);
1304 	dqvp = ump->um_quotas[type];
1305 	if (dqvp == NULLVP || (ump->um_qflags[type] & QTF_CLOSING)) {
1306 		*dqp = NODQUOT;
1307 		UFS_UNLOCK(ump);
1308 		return (EINVAL);
1309 	}
1310 	vref(dqvp);
1311 	UFS_UNLOCK(ump);
1312 	error = 0;
1313 	dqvplocked = 0;
1314 
1315 	/*
1316 	 * Check the cache first.
1317 	 */
1318 	dqh = DQHASH(dqvp, id);
1319 	DQH_LOCK();
1320 	dq = dqhashfind(dqh, id, dqvp);
1321 	if (dq != NULL) {
1322 		DQH_UNLOCK();
1323 hfound:		DQI_LOCK(dq);
1324 		DQI_WAIT(dq, PINOD+1, "dqget");
1325 		DQI_UNLOCK(dq);
1326 		if (dq->dq_ump == NULL) {
1327 			dqrele(vp, dq);
1328 			dq = NODQUOT;
1329 			error = EIO;
1330 		}
1331 		*dqp = dq;
1332 		if (dqvplocked)
1333 			vput(dqvp);
1334 		else
1335 			vrele(dqvp);
1336 		return (error);
1337 	}
1338 
1339 	/*
1340 	 * Quota vnode lock is before DQ_LOCK. Acquire dqvp lock there
1341 	 * since new dq will appear on the hash chain DQ_LOCKed.
1342 	 */
1343 	if (vp != dqvp) {
1344 		DQH_UNLOCK();
1345 		vn_lock(dqvp, LK_SHARED | LK_RETRY);
1346 		dqvplocked = 1;
1347 		DQH_LOCK();
1348 		/*
1349 		 * Recheck the cache after sleep for quota vnode lock.
1350 		 */
1351 		dq = dqhashfind(dqh, id, dqvp);
1352 		if (dq != NULL) {
1353 			DQH_UNLOCK();
1354 			goto hfound;
1355 		}
1356 	}
1357 
1358 	/*
1359 	 * Not in cache, allocate a new one or take it from the
1360 	 * free list.
1361 	 */
1362 	if (TAILQ_FIRST(&dqfreelist) == NODQUOT &&
1363 	    numdquot < MAXQUOTAS * desiredvnodes)
1364 		desireddquot += DQUOTINC;
1365 	if (numdquot < desireddquot) {
1366 		numdquot++;
1367 		DQH_UNLOCK();
1368 		dq1 = malloc(sizeof *dq1, M_DQUOT, M_WAITOK | M_ZERO);
1369 		mtx_init(&dq1->dq_lock, "dqlock", NULL, MTX_DEF);
1370 		DQH_LOCK();
1371 		/*
1372 		 * Recheck the cache after sleep for memory.
1373 		 */
1374 		dq = dqhashfind(dqh, id, dqvp);
1375 		if (dq != NULL) {
1376 			numdquot--;
1377 			DQH_UNLOCK();
1378 			mtx_destroy(&dq1->dq_lock);
1379 			free(dq1, M_DQUOT);
1380 			goto hfound;
1381 		}
1382 		dq = dq1;
1383 	} else {
1384 		if ((dq = TAILQ_FIRST(&dqfreelist)) == NULL) {
1385 			DQH_UNLOCK();
1386 			tablefull("dquot");
1387 			*dqp = NODQUOT;
1388 			if (dqvplocked)
1389 				vput(dqvp);
1390 			else
1391 				vrele(dqvp);
1392 			return (EUSERS);
1393 		}
1394 		if (dq->dq_cnt || (dq->dq_flags & DQ_MOD))
1395 			panic("dqget: free dquot isn't %p", dq);
1396 		TAILQ_REMOVE(&dqfreelist, dq, dq_freelist);
1397 		if (dq->dq_ump != NULL)
1398 			LIST_REMOVE(dq, dq_hash);
1399 	}
1400 
1401 	/*
1402 	 * Dq is put into hash already locked to prevent parallel
1403 	 * usage while it is being read from file.
1404 	 */
1405 	dq->dq_flags = DQ_LOCK;
1406 	dq->dq_id = id;
1407 	dq->dq_type = type;
1408 	dq->dq_ump = ump;
1409 	LIST_INSERT_HEAD(dqh, dq, dq_hash);
1410 	DQREF(dq);
1411 	DQH_UNLOCK();
1412 
1413 	/*
1414 	 * Read the requested quota record from the quota file, performing
1415 	 * any necessary conversions.
1416 	 */
1417 	if (ump->um_qflags[type] & QTF_64BIT) {
1418 		recsize = sizeof(struct dqblk64);
1419 		base = sizeof(struct dqhdr64);
1420 	} else {
1421 		recsize = sizeof(struct dqblk32);
1422 		base = 0;
1423 	}
1424 	auio.uio_iov = &aiov;
1425 	auio.uio_iovcnt = 1;
1426 	aiov.iov_base = buf;
1427 	aiov.iov_len = recsize;
1428 	auio.uio_resid = recsize;
1429 	auio.uio_offset = base + id * recsize;
1430 	auio.uio_segflg = UIO_SYSSPACE;
1431 	auio.uio_rw = UIO_READ;
1432 	auio.uio_td = (struct thread *)0;
1433 
1434 	error = VOP_READ(dqvp, &auio, 0, ump->um_cred[type]);
1435 	if (auio.uio_resid == recsize && error == 0) {
1436 		bzero(&dq->dq_dqb, sizeof(dq->dq_dqb));
1437 	} else {
1438 		if (ump->um_qflags[type] & QTF_64BIT)
1439 			dqb64_dq((struct dqblk64 *)buf, dq);
1440 		else
1441 			dqb32_dq((struct dqblk32 *)buf, dq);
1442 	}
1443 	if (dqvplocked)
1444 		vput(dqvp);
1445 	else
1446 		vrele(dqvp);
1447 	/*
1448 	 * I/O error in reading quota file, release
1449 	 * quota structure and reflect problem to caller.
1450 	 */
1451 	if (error) {
1452 		DQH_LOCK();
1453 		dq->dq_ump = NULL;
1454 		LIST_REMOVE(dq, dq_hash);
1455 		DQH_UNLOCK();
1456 		DQI_LOCK(dq);
1457 		if (dq->dq_flags & DQ_WANT)
1458 			wakeup(dq);
1459 		dq->dq_flags = 0;
1460 		DQI_UNLOCK(dq);
1461 		dqrele(vp, dq);
1462 		*dqp = NODQUOT;
1463 		return (error);
1464 	}
1465 	DQI_LOCK(dq);
1466 	/*
1467 	 * Check for no limit to enforce.
1468 	 * Initialize time values if necessary.
1469 	 */
1470 	if (dq->dq_isoftlimit == 0 && dq->dq_bsoftlimit == 0 &&
1471 	    dq->dq_ihardlimit == 0 && dq->dq_bhardlimit == 0)
1472 		dq->dq_flags |= DQ_FAKE;
1473 	if (dq->dq_id != 0) {
1474 		if (dq->dq_btime == 0) {
1475 			dq->dq_btime = time_second + ump->um_btime[type];
1476 			if (dq->dq_bsoftlimit &&
1477 			    dq->dq_curblocks >= dq->dq_bsoftlimit)
1478 				dq->dq_flags |= DQ_MOD;
1479 		}
1480 		if (dq->dq_itime == 0) {
1481 			dq->dq_itime = time_second + ump->um_itime[type];
1482 			if (dq->dq_isoftlimit &&
1483 			    dq->dq_curinodes >= dq->dq_isoftlimit)
1484 				dq->dq_flags |= DQ_MOD;
1485 		}
1486 	}
1487 	DQI_WAKEUP(dq);
1488 	DQI_UNLOCK(dq);
1489 	*dqp = dq;
1490 	return (0);
1491 }
1492 
1493 #ifdef DIAGNOSTIC
1494 /*
1495  * Obtain a reference to a dquot.
1496  */
1497 static void
1498 dqref(struct dquot *dq)
1499 {
1500 
1501 	dq->dq_cnt++;
1502 }
1503 #endif
1504 
1505 /*
1506  * Release a reference to a dquot.
1507  */
1508 void
1509 dqrele(struct vnode *vp, struct dquot *dq)
1510 {
1511 
1512 	if (dq == NODQUOT)
1513 		return;
1514 	DQH_LOCK();
1515 	KASSERT(dq->dq_cnt > 0, ("Lost dq %p reference 1", dq));
1516 	if (dq->dq_cnt > 1) {
1517 		dq->dq_cnt--;
1518 		DQH_UNLOCK();
1519 		return;
1520 	}
1521 	DQH_UNLOCK();
1522 sync:
1523 	(void) dqsync(vp, dq);
1524 
1525 	DQH_LOCK();
1526 	KASSERT(dq->dq_cnt > 0, ("Lost dq %p reference 2", dq));
1527 	if (--dq->dq_cnt > 0)
1528 	{
1529 		DQH_UNLOCK();
1530 		return;
1531 	}
1532 
1533 	/*
1534 	 * The dq may become dirty after it is synced but before it is
1535 	 * put to the free list. Checking the DQ_MOD there without
1536 	 * locking dq should be safe since no other references to the
1537 	 * dq exist.
1538 	 */
1539 	if ((dq->dq_flags & DQ_MOD) != 0) {
1540 		dq->dq_cnt++;
1541 		DQH_UNLOCK();
1542 		goto sync;
1543 	}
1544 	TAILQ_INSERT_TAIL(&dqfreelist, dq, dq_freelist);
1545 	DQH_UNLOCK();
1546 }
1547 
1548 /*
1549  * Update the disk quota in the quota file.
1550  */
1551 static int
1552 dqsync(struct vnode *vp, struct dquot *dq)
1553 {
1554 	uint8_t buf[sizeof(struct dqblk64)];
1555 	off_t base, recsize;
1556 	struct vnode *dqvp;
1557 	struct iovec aiov;
1558 	struct uio auio;
1559 	int error;
1560 	struct mount *mp;
1561 	struct ufsmount *ump;
1562 
1563 #ifdef DEBUG_VFS_LOCKS
1564 	if (vp != NULL)
1565 		ASSERT_VOP_ELOCKED(vp, "dqsync");
1566 #endif
1567 
1568 	mp = NULL;
1569 	error = 0;
1570 	if (dq == NODQUOT)
1571 		panic("dqsync: dquot");
1572 	if ((ump = dq->dq_ump) == NULL)
1573 		return (0);
1574 	UFS_LOCK(ump);
1575 	if ((dqvp = ump->um_quotas[dq->dq_type]) == NULLVP) {
1576 		if (vp == NULL) {
1577 			UFS_UNLOCK(ump);
1578 			return (0);
1579 		} else
1580 			panic("dqsync: file");
1581 	}
1582 	vref(dqvp);
1583 	UFS_UNLOCK(ump);
1584 
1585 	DQI_LOCK(dq);
1586 	if ((dq->dq_flags & DQ_MOD) == 0) {
1587 		DQI_UNLOCK(dq);
1588 		vrele(dqvp);
1589 		return (0);
1590 	}
1591 	DQI_UNLOCK(dq);
1592 
1593 	(void) vn_start_secondary_write(dqvp, &mp, V_WAIT);
1594 	if (vp != dqvp)
1595 		vn_lock(dqvp, LK_EXCLUSIVE | LK_RETRY);
1596 
1597 	DQI_LOCK(dq);
1598 	DQI_WAIT(dq, PINOD+2, "dqsync");
1599 	if ((dq->dq_flags & DQ_MOD) == 0)
1600 		goto out;
1601 	dq->dq_flags |= DQ_LOCK;
1602 	DQI_UNLOCK(dq);
1603 
1604 	/*
1605 	 * Write the quota record to the quota file, performing any
1606 	 * necessary conversions.  See dqget() for additional details.
1607 	 */
1608 	if (ump->um_qflags[dq->dq_type] & QTF_64BIT) {
1609 		dq_dqb64(dq, (struct dqblk64 *)buf);
1610 		recsize = sizeof(struct dqblk64);
1611 		base = sizeof(struct dqhdr64);
1612 	} else {
1613 		dq_dqb32(dq, (struct dqblk32 *)buf);
1614 		recsize = sizeof(struct dqblk32);
1615 		base = 0;
1616 	}
1617 
1618 	auio.uio_iov = &aiov;
1619 	auio.uio_iovcnt = 1;
1620 	aiov.iov_base = buf;
1621 	aiov.iov_len = recsize;
1622 	auio.uio_resid = recsize;
1623 	auio.uio_offset = base + dq->dq_id * recsize;
1624 	auio.uio_segflg = UIO_SYSSPACE;
1625 	auio.uio_rw = UIO_WRITE;
1626 	auio.uio_td = (struct thread *)0;
1627 	error = VOP_WRITE(dqvp, &auio, 0, dq->dq_ump->um_cred[dq->dq_type]);
1628 	if (auio.uio_resid && error == 0)
1629 		error = EIO;
1630 
1631 	DQI_LOCK(dq);
1632 	DQI_WAKEUP(dq);
1633 	dq->dq_flags &= ~DQ_MOD;
1634 out:
1635 	DQI_UNLOCK(dq);
1636 	if (vp != dqvp)
1637 		vput(dqvp);
1638 	else
1639 		vrele(dqvp);
1640 	vn_finished_secondary_write(mp);
1641 	return (error);
1642 }
1643 
1644 /*
1645  * Flush all entries from the cache for a particular vnode.
1646  */
1647 static int
1648 dqflush(struct vnode *vp)
1649 {
1650 	struct dquot *dq, *nextdq;
1651 	struct dqhash *dqh;
1652 	int error;
1653 
1654 	/*
1655 	 * Move all dquot's that used to refer to this quota
1656 	 * file off their hash chains (they will eventually
1657 	 * fall off the head of the free list and be re-used).
1658 	 */
1659 	error = 0;
1660 	DQH_LOCK();
1661 	for (dqh = &dqhashtbl[dqhash]; dqh >= dqhashtbl; dqh--) {
1662 		for (dq = LIST_FIRST(dqh); dq; dq = nextdq) {
1663 			nextdq = LIST_NEXT(dq, dq_hash);
1664 			if (dq->dq_ump->um_quotas[dq->dq_type] != vp)
1665 				continue;
1666 			if (dq->dq_cnt)
1667 				error = EBUSY;
1668 			else {
1669 				LIST_REMOVE(dq, dq_hash);
1670 				dq->dq_ump = NULL;
1671 			}
1672 		}
1673 	}
1674 	DQH_UNLOCK();
1675 	return (error);
1676 }
1677 
1678 /*
1679  * The following three functions are provided for the adjustment of
1680  * quotas by the soft updates code.
1681  */
1682 #ifdef SOFTUPDATES
1683 /*
1684  * Acquire a reference to the quota structures associated with a vnode.
1685  * Return count of number of quota structures found.
1686  */
1687 int
1688 quotaref(vp, qrp)
1689 	struct vnode *vp;
1690 	struct dquot **qrp;
1691 {
1692 	struct inode *ip;
1693 	struct dquot *dq;
1694 	int i, found;
1695 
1696 	for (i = 0; i < MAXQUOTAS; i++)
1697 		qrp[i] = NODQUOT;
1698 	/*
1699 	 * Disk quotas must be turned off for system files.  Currently
1700 	 * snapshot and quota files.
1701 	 */
1702 	if ((vp->v_vflag & VV_SYSTEM) != 0)
1703 		return (0);
1704 	/*
1705 	 * Iterate through and copy active quotas.
1706 	 */
1707 	found = 0;
1708 	ip = VTOI(vp);
1709 	mtx_lock(&dqhlock);
1710 	for (i = 0; i < MAXQUOTAS; i++) {
1711 		if ((dq = ip->i_dquot[i]) == NODQUOT)
1712 			continue;
1713 		DQREF(dq);
1714 		qrp[i] = dq;
1715 		found++;
1716 	}
1717 	mtx_unlock(&dqhlock);
1718 	return (found);
1719 }
1720 
1721 /*
1722  * Release a set of quota structures obtained from a vnode.
1723  */
1724 void
1725 quotarele(qrp)
1726 	struct dquot **qrp;
1727 {
1728 	struct dquot *dq;
1729 	int i;
1730 
1731 	for (i = 0; i < MAXQUOTAS; i++) {
1732 		if ((dq = qrp[i]) == NODQUOT)
1733 			continue;
1734 		dqrele(NULL, dq);
1735 	}
1736 }
1737 
1738 /*
1739  * Adjust the number of blocks associated with a quota.
1740  * Positive numbers when adding blocks; negative numbers when freeing blocks.
1741  */
1742 void
1743 quotaadj(qrp, ump, blkcount)
1744 	struct dquot **qrp;
1745 	struct ufsmount *ump;
1746 	int64_t blkcount;
1747 {
1748 	struct dquot *dq;
1749 	ufs2_daddr_t ncurblocks;
1750 	int i;
1751 
1752 	if (blkcount == 0)
1753 		return;
1754 	for (i = 0; i < MAXQUOTAS; i++) {
1755 		if ((dq = qrp[i]) == NODQUOT)
1756 			continue;
1757 		DQI_LOCK(dq);
1758 		DQI_WAIT(dq, PINOD+1, "adjqta");
1759 		ncurblocks = dq->dq_curblocks + blkcount;
1760 		if (ncurblocks >= 0)
1761 			dq->dq_curblocks = ncurblocks;
1762 		else
1763 			dq->dq_curblocks = 0;
1764 		if (blkcount < 0)
1765 			dq->dq_flags &= ~DQ_BLKS;
1766 		else if (dq->dq_curblocks + blkcount >= dq->dq_bsoftlimit &&
1767 			 dq->dq_curblocks < dq->dq_bsoftlimit)
1768 			dq->dq_btime = time_second + ump->um_btime[i];
1769 		dq->dq_flags |= DQ_MOD;
1770 		DQI_UNLOCK(dq);
1771 	}
1772 }
1773 #endif /* SOFTUPDATES */
1774 
1775 /*
1776  * 32-bit / 64-bit conversion functions.
1777  *
1778  * 32-bit quota records are stored in native byte order.  Attention must
1779  * be paid to overflow issues.
1780  *
1781  * 64-bit quota records are stored in network byte order.
1782  */
1783 
1784 #define CLIP32(u64) (u64 > UINT32_MAX ? UINT32_MAX : (uint32_t)u64)
1785 
1786 /*
1787  * Convert 32-bit host-order structure to dquot.
1788  */
1789 static void
1790 dqb32_dq(const struct dqblk32 *dqb32, struct dquot *dq)
1791 {
1792 
1793 	dq->dq_bhardlimit = dqb32->dqb_bhardlimit;
1794 	dq->dq_bsoftlimit = dqb32->dqb_bsoftlimit;
1795 	dq->dq_curblocks = dqb32->dqb_curblocks;
1796 	dq->dq_ihardlimit = dqb32->dqb_ihardlimit;
1797 	dq->dq_isoftlimit = dqb32->dqb_isoftlimit;
1798 	dq->dq_curinodes = dqb32->dqb_curinodes;
1799 	dq->dq_btime = dqb32->dqb_btime;
1800 	dq->dq_itime = dqb32->dqb_itime;
1801 }
1802 
1803 /*
1804  * Convert 64-bit network-order structure to dquot.
1805  */
1806 static void
1807 dqb64_dq(const struct dqblk64 *dqb64, struct dquot *dq)
1808 {
1809 
1810 	dq->dq_bhardlimit = be64toh(dqb64->dqb_bhardlimit);
1811 	dq->dq_bsoftlimit = be64toh(dqb64->dqb_bsoftlimit);
1812 	dq->dq_curblocks = be64toh(dqb64->dqb_curblocks);
1813 	dq->dq_ihardlimit = be64toh(dqb64->dqb_ihardlimit);
1814 	dq->dq_isoftlimit = be64toh(dqb64->dqb_isoftlimit);
1815 	dq->dq_curinodes = be64toh(dqb64->dqb_curinodes);
1816 	dq->dq_btime = be64toh(dqb64->dqb_btime);
1817 	dq->dq_itime = be64toh(dqb64->dqb_itime);
1818 }
1819 
1820 /*
1821  * Convert dquot to 32-bit host-order structure.
1822  */
1823 static void
1824 dq_dqb32(const struct dquot *dq, struct dqblk32 *dqb32)
1825 {
1826 
1827 	dqb32->dqb_bhardlimit = CLIP32(dq->dq_bhardlimit);
1828 	dqb32->dqb_bsoftlimit = CLIP32(dq->dq_bsoftlimit);
1829 	dqb32->dqb_curblocks = CLIP32(dq->dq_curblocks);
1830 	dqb32->dqb_ihardlimit = CLIP32(dq->dq_ihardlimit);
1831 	dqb32->dqb_isoftlimit = CLIP32(dq->dq_isoftlimit);
1832 	dqb32->dqb_curinodes = CLIP32(dq->dq_curinodes);
1833 	dqb32->dqb_btime = CLIP32(dq->dq_btime);
1834 	dqb32->dqb_itime = CLIP32(dq->dq_itime);
1835 }
1836 
1837 /*
1838  * Convert dquot to 64-bit network-order structure.
1839  */
1840 static void
1841 dq_dqb64(const struct dquot *dq, struct dqblk64 *dqb64)
1842 {
1843 
1844 	dqb64->dqb_bhardlimit = htobe64(dq->dq_bhardlimit);
1845 	dqb64->dqb_bsoftlimit = htobe64(dq->dq_bsoftlimit);
1846 	dqb64->dqb_curblocks = htobe64(dq->dq_curblocks);
1847 	dqb64->dqb_ihardlimit = htobe64(dq->dq_ihardlimit);
1848 	dqb64->dqb_isoftlimit = htobe64(dq->dq_isoftlimit);
1849 	dqb64->dqb_curinodes = htobe64(dq->dq_curinodes);
1850 	dqb64->dqb_btime = htobe64(dq->dq_btime);
1851 	dqb64->dqb_itime = htobe64(dq->dq_itime);
1852 }
1853 
1854 /*
1855  * Convert 64-bit host-order structure to 32-bit host-order structure.
1856  */
1857 static void
1858 dqb64_dqb32(const struct dqblk64 *dqb64, struct dqblk32 *dqb32)
1859 {
1860 
1861 	dqb32->dqb_bhardlimit = CLIP32(dqb64->dqb_bhardlimit);
1862 	dqb32->dqb_bsoftlimit = CLIP32(dqb64->dqb_bsoftlimit);
1863 	dqb32->dqb_curblocks = CLIP32(dqb64->dqb_curblocks);
1864 	dqb32->dqb_ihardlimit = CLIP32(dqb64->dqb_ihardlimit);
1865 	dqb32->dqb_isoftlimit = CLIP32(dqb64->dqb_isoftlimit);
1866 	dqb32->dqb_curinodes = CLIP32(dqb64->dqb_curinodes);
1867 	dqb32->dqb_btime = CLIP32(dqb64->dqb_btime);
1868 	dqb32->dqb_itime = CLIP32(dqb64->dqb_itime);
1869 }
1870 
1871 /*
1872  * Convert 32-bit host-order structure to 64-bit host-order structure.
1873  */
1874 static void
1875 dqb32_dqb64(const struct dqblk32 *dqb32, struct dqblk64 *dqb64)
1876 {
1877 
1878 	dqb64->dqb_bhardlimit = dqb32->dqb_bhardlimit;
1879 	dqb64->dqb_bsoftlimit = dqb32->dqb_bsoftlimit;
1880 	dqb64->dqb_curblocks = dqb32->dqb_curblocks;
1881 	dqb64->dqb_ihardlimit = dqb32->dqb_ihardlimit;
1882 	dqb64->dqb_isoftlimit = dqb32->dqb_isoftlimit;
1883 	dqb64->dqb_curinodes = dqb32->dqb_curinodes;
1884 	dqb64->dqb_btime = dqb32->dqb_btime;
1885 	dqb64->dqb_itime = dqb32->dqb_itime;
1886 }
1887